Reinforcement Learning-Based Differential Evolution With Cooperative Coevolution for a Compensatory Neuro-Fuzzy...

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Computational Intelligence

Background:

• Compensatory neuro-fuzzy controllers (CNFCs) offer enhanced adaptability for control problems.
• Optimizing controller parameters is crucial for effective nonlinear system management.
• Traditional optimization methods may struggle with complex, nonlinear control tasks.

Purpose of the Study:

• To integrate reinforcement learning-based differential evolution (DE) with cooperative coevolution (R-CCDE) for CNFC parameter optimization.
• To develop an advanced control policy for nonlinear systems using the proposed R-CCDE method.
• To evaluate the performance of the R-CCDE method against existing DE techniques.

Main Methods:

• The R-CCDE method was employed to evolve populations and adjust CNFC parameters.
• Cooperative coevolution was utilized within the DE framework for parameter optimization.
• A reinforcement signal derived from the R-CCDE fitness function guided controller selection.

Main Results:

• The R-CCDE method successfully identified an optimal controller for nonlinear system problems.
• Simulation results indicated the superiority of the R-CCDE approach over various DE methods.
• The proposed R-CCDE method achieved enhanced adaptability and effectiveness in control applications.

Conclusions:

• The R-CCDE method provides a robust and effective approach for optimizing CNFCs.
• This integration offers significant improvements for tackling complex nonlinear control challenges.
• The study highlights the potential of R-CCDE in advanced control system design.